A distribution network cable fault diagnosis method, device, equipment and storage medium

Abstract

The application relates to the technical field of cable fault diagnosis, and discloses a distribution network cable fault diagnosis method, device, equipment and storage medium. The method comprises the following steps: obtaining a target current signal of a target distribution network cable to be diagnosed; initializing a frequency scaling factor and a center frequency of a Morlet wavelet base function based on a target sampling frequency, and constructing a frequency scaling kernel function; performing one-dimensional convolution operation on the target current signal based on the frequency scaling kernel function, and generating a fault feature map; processing the fault feature map to obtain a fusion feature map; inputting the fusion feature map into a target connection linear layer for processing, and determining a fault type; determining a fault point position based on a time difference of a fault harmonic peak value of adjacent collection points and a current wave propagation speed; determining a fault target investigation path based on the fault type, the fault point position and a distribution network topological structure, and investigating and determining a target fault. The scheme optimizes a fault diagnosis network structure, and improves fault positioning precision and fault recognition accuracy.

Description

Technical Field

[0001] This invention belongs to the field of cable fault diagnosis technology, specifically relating to methods, devices, equipment and storage media for fault diagnosis of distribution network cables. Background Technology

[0002] As the core transmission carrier of medium-voltage distribution networks, the fault diagnosis and maintenance technology of distribution cables has evolved from traditional manual inspection to data-driven intelligent diagnosis. Current traditional fault diagnosis technologies primarily employ impedance methods, traveling wave methods, and manual inspection combined with partial discharge detection. Impedance methods locate faults by measuring the impedance difference before and after the fault point, requiring precise cable parameters (such as length and wave impedance), and have a low recognition rate for high-resistance faults. Traveling wave methods utilize the propagation characteristics of transient traveling waves to locate faults, but they require extremely high synchronization of transient signal acquisition and are easily affected by reflections from cable branches and joints. Manual inspection combined with partial discharge detection relies on the experience of maintenance personnel, requiring segment-by-segment inspection, which is inefficient and unsuitable for the rapid fault location needs of long-distance, complex laying scenarios. With the development of artificial intelligence technology, diagnostic methods based on models such as convolutional neural networks and long short-term memory networks have emerged. These methods typically collect multi-dimensional signals such as cable current, vibration, and temperature, preprocess them, and then input them into a deep network to achieve fault classification and location. Among them, the closest existing technology to this invention is the deep learning method for diagnosing distribution network cable faults, whose core steps include: deployment of multiple types of sensors, signal acquisition, complex signal preprocessing, deep network training and fault classification, and fault location by combining a simple distance formula.

[0003] Current technological development trends aim to improve diagnostic accuracy, but most focus on increasing model complexity or multi-signal fusion, neglecting the actual needs of distribution network operation and maintenance. Existing fault diagnosis models are bulky, require high computing power, are unsuitable for portable field devices, have low fault location accuracy, lack efficient troubleshooting path planning, resulting in low operation and maintenance efficiency, inaccurate utilization of fault characteristics, and limited diagnostic accuracy. Summary of the Invention

[0004] In view of the shortcomings of the prior art, the purpose of this invention is to provide a method, device, equipment and medium for diagnosing faults in distribution network cables, so as to solve the problems of bulky structure, low positioning accuracy and low identification accuracy of the prior art.

[0005] According to one aspect of this application, a method for diagnosing faults in distribution network cables is disclosed, the method comprising: Acquire the target current signal of the distribution network cable to be diagnosed; Based on the target sampling frequency of the target current signal, initialize the frequency scaling factor and center frequency of the Morlet wavelet basis function, and construct the frequency scaling kernel function; The target current signal is subjected to a one-dimensional convolution operation based on the frequency scaling kernel function to generate a multi-channel fault feature map. The fault feature map is processed to obtain a fused feature map; The fused feature map is input into a target connected linear layer for processing to determine the fault type corresponding to the fused feature map. The dimension of the weight matrix of the target connected linear layer is configured to be associated with the frequency channel dimension of the fused feature map, without changing with the time point of the fused feature map. The location of the fault point is determined based on the time difference of the peak fault harmonics and the propagation speed of the current wave at adjacent acquisition points in the fault feature map. Based on the fault type, the location of the fault point, and the distribution network topology, a fault target investigation path is determined, and the target distribution network cable to be diagnosed is investigated based on the fault target investigation path to diagnose the target fault of the target distribution network cable.

[0006] In some embodiments, acquiring the target current signal of the target distribution network cable to be diagnosed includes: Obtain the preset sampling frequency and preset sampling duration; Based on the preset sampling frequency and the preset sampling duration, the number of data points of the target distribution network cable to be diagnosed is determined; At a selected node of the target distribution network cable to be diagnosed, the original current signal of the target distribution network cable to be diagnosed is collected by a broadband current sensor for the preset sampling duration; Synchronously record associated metadata during the acquisition process, including at least the acquisition timestamp, sensor location identifier, and cable operating load; The data points of the acquired raw current signal are logically bound to the associated metadata in a predetermined format to generate and store the raw current signal vector. The original current signal vector is preprocessed to obtain the target current signal.

[0007] In some embodiments, the process of initializing the frequency scaling factor and center frequency of the Morlet wavelet basis function based on the target sampling frequency of the target current signal and constructing the frequency scaling kernel function includes: Obtain the target sampling frequency of the target current signal; Based on the target sampling frequency, determine the kernel center frequency and the range of time sample points for the Morlet wavelet basis function; Based on the core harmonic frequency band of the target type of cable fault, set a set of initial trainable scaling factors. The frequency scaling kernel function is constructed using the kernel center frequency, the time sample points, and the initialized trainable scaling factor; Among them, the core center frequency ; In the formula, For the core center frequency, The core center frequency is in frequency form. The target sampling frequency for the target current signal; The range of values ​​for time sample points is ; The expression for the frequency scaling kernel function is: ; In the formula, : Frequency scaling kernel output value; Time sample points, from the first step Confirmed, covering one complete fault transient cycle; : Trainable scaling factor, the initial trainable scaling factor is , The core channel number is used to match the target type of cable fault in the core harmonic frequency band; Bandwidth constant.

[0008] Specifically, in this scheme, the operation of constructing the frequency scaling kernel function corresponds to a one-dimensional convolutional layer, whose input and output dimensions and functions are as follows: The input is the target current signal, with dimensions [U, 1], meaning the input is a single-channel current signal with a signal length of U; the output is a multi-channel frequency scaling kernel function, with dimensions [L, C], where L is the kernel function length, which is related to the range of time sample points. The number of sampling points is consistent, by It is determined that L×sampling interval = 4π / This means covering a complete fault transient cycle. C is the number of core channels, which is consistent with the number of core harmonic frequency bands of the cable fault. For example, when there are 6 core harmonic frequency bands, C=6.

[0009] Furthermore, the output of the frequency scaling kernel function is subjected to a series of convolution operations with the preprocessed current signal. Connect and generate fault feature maps Its convolutional budget logic is as follows: Input 1: The dimension is [L,C], and the input is the preprocessed current signal. Dimensions [N, 1], Output: Fault Feature Map The dimension is [P, C], where P is the length of the feature map, derived from the formula... calculate.

[0010] Furthermore, Each channel, for current signal Perform locally weighted convolution, i.e., the kernel at the j-th time sample point. and The Multiplying the sample points together and summing them up, we get the nth sample point. The first time window, the... The characteristic values ​​of the channel.

[0011] Furthermore, batch normalization is performed. The feature values ​​of each time window are normalized to ensure that the feature amplitudes of different channels are consistent, thus avoiding imbalance in weight updates during training.

[0012] Furthermore, fault characteristic diagrams It is a multi-channel, single-time-dimensional feature representation. The final fused feature map can be generated through time-frequency fusion, and its dimension is typically [missing information]. ,in, This represents the length of the time axis after downsampling. This represents the number of channels after merging.

[0013] In some instances, the one-dimensional convolution operation on the target current signal based on the frequency scaling kernel function generates a multi-channel fault feature map, with a convolution stride of... , The expression for one-dimensional convolution is: ; In the formula: : Convolution stride; Fault feature map output by convolution, dimension , For the number of channels, The length of the feature map is given, and each channel corresponds to a type of current fault harmonic feature. : Feature map time axis index, value This corresponds to the time window after downsampling; The kernel function length is matched with the kernel parameters in the second step to ensure the capture of transient local features of the current. : No. Each time sample point, scaling factor The kernel value comes from the initialization result in the second step and is optimized with fault characteristics in subsequent training; The amplitude of the preprocessed current signal at the corresponding index comes from step two. ; Batch normalization avoids imbalance in weight updates caused by differences in current amplitude; Preprocessed current signal The total number of sampling points; Feature map length The calculation formula is: .

[0014] In some embodiments, processing the fault feature map to obtain a fused feature map includes: The fault feature map is subjected to one-dimensional convolution and pooling to obtain a time-domain refined feature map; The temporal refined feature map is reshaped into a two-dimensional tensor, and a two-dimensional convolution operation is performed on the two-dimensional tensor to obtain the fused feature map.

[0015] In some embodiments, the step of inputting the fused feature map into a target connected linear layer for processing to determine the fault type corresponding to the fused feature map includes: After the fused feature map is input into the target connected linear layer, classification scoring is performed based on the scoring formula to obtain the fault classification scores corresponding to each of the multiple fault types in the fused feature map. The scores of multiple fault classifications are normalized to obtain the fault type corresponding to the fused feature map. The scoring formula is as follows: ; In the formula, : No. Classification scores for different types of faults ; : Feature map channel index, fixed at 1, from the single-channel output of the fused feature map; Time axis index, value ; : Length of the time axis after downsampling; Frequency channel index, value ; : The transpose of the matrix in The transpose of the fused feature map whose position is determined by the elements; Target connection weights of the linear layer, dimension ; : No. Fault bias term.

[0016] In some embodiments, determining the fault location based on the time difference of fault harmonic peak values ​​and current wave propagation velocity between adjacent acquisition points in the fault feature map includes: Extract the peak times of the first harmonic current fault and the peak times of the second harmonic current fault from adjacent acquisition points in the fault feature map. Determine the current wave propagation speed based on the type of the target distribution network cable to be diagnosed; Based on the peak times of the first harmonic of the current fault, the peak times of the second harmonic of the current fault, and the propagation speed of the current wave, the location of the fault point is determined using a distance formula, wherein the distance formula is: ; In the formula: : The distance from the fault point to the starting data collection point; Spacing between data collection points; : Speed ​​of propagation of electric current waves; : The peak time of the first harmonic; : The peak time of the second harmonic; The absolute value of the time difference between the peak time of the first harmonic and the peak time of the second harmonic.

[0017] According to another aspect of this application, a fault diagnosis device for distribution network cables is also disclosed, the device comprising: The target current signal acquisition module is used to acquire the target current signal of the target distribution network cable to be diagnosed. The frequency scaling kernel function construction module is used to initialize the frequency scaling factor and center frequency of the Morlet wavelet basis function based on the target sampling frequency of the target current signal, and construct the frequency scaling kernel function. The fault feature map generation module is used to perform a one-dimensional convolution operation on the target current signal based on the frequency scaling kernel function to generate a multi-channel fault feature map. The fusion feature map determination module is used to process the fault feature map to obtain the fusion feature map; The fault type determination module is used to input the fused feature map into the target connected linear layer for processing in order to determine the fault type corresponding to the fused feature map. The dimension of the weight matrix of the target connected linear layer is configured to be associated with the frequency channel dimension of the fused feature map, and does not change with the time point of the fused feature map. The fault location determination module determines the fault location based on the time difference of the fault harmonic peak values ​​of adjacent acquisition points and the current wave propagation speed in the fault feature map. The target fault diagnosis module is used to determine the fault target investigation path based on the fault type, the fault point location and the distribution network topology, and to investigate the target distribution network cable to be diagnosed based on the fault target investigation path, so as to diagnose the target fault of the target distribution network cable to be diagnosed.

[0018] According to another aspect of this application, an electronic device is also disclosed, the electronic device including a memory and at least one processor, the memory storing instructions; the at least one processor invokes the instructions in the memory to cause the electronic device to perform the steps of the distribution network cable fault diagnosis method as described in any of the preceding claims.

[0019] According to another aspect of this application, a computer-readable storage medium is also disclosed, wherein instructions are stored on the computer-readable storage medium, characterized in that, when executed by a processor, the instructions implement the various steps of the distribution network cable fault diagnosis method as described in any of the preceding claims.

[0020] The present invention includes, but is not limited to, the following beneficial effects: (1) This solution obtains the target current signal of the target distribution network cable to be diagnosed; based on the target sampling frequency, initializes the frequency scaling factor and center frequency of the Morlet wavelet basis function, and constructs the frequency scaling kernel function; performs one-dimensional convolution operation on the target current signal based on the frequency scaling kernel function to generate a fault feature map; processes the fault feature map to obtain a fused feature map; inputs the fused feature map into the target connected linear layer for processing to determine the fault type; determines the fault location based on the fault harmonic peak time difference and current wave propagation speed of adjacent acquisition points; determines the fault target investigation path based on the fault type, fault location, and distribution network topology, and investigates and determines the steps of the target fault, optimizes the fault diagnosis network structure, reduces the number of trainable parameters, reduces the requirements for computing power, and makes the model adaptable to portable equipment on site; improves the fault location accuracy, generates the optimal investigation path based on the distribution network topology data, and improves the operation and maintenance efficiency; avoids relying on complex time-frequency transformation for fault feature extraction, and improves the efficiency of operation and maintenance. (1) The identification rate of minor faults; (2) The signal acquisition stage of this scheme uses a wideband current sensor to capture power frequency and low frequency harmonic current and medium frequency transient current. It does not require the deployment of high frequency sensors or multiple types of sensors, which greatly reduces the equipment procurement and installation costs. At the same time, the acquired signal focuses on the core features of the fault, which has stronger anti-electromagnetic interference capability and significantly improved field compatibility; (3) A lightweight network structure is constructed by using a dedicated frequency scaling convolutional layer and a partially connected linear layer, which greatly reduces the number of trainable parameters, reduces the demand for computing power, and makes the model inference faster. It can be adapted to portable diagnostic equipment on site and meet the real-time diagnostic needs; (4) The fault location algorithm is optimized by combining the current wave propagation characteristics and distribution network topology data to achieve high-precision location and generate target investigation paths simultaneously, which shortens the fault investigation time and greatly improves the efficiency of distribution network operation and maintenance; (5) This scheme mines fault features through frequency scaling convolution, which can effectively retain key information. It not only improves the diagnostic accuracy of multiple types of faults, but also enhances the ability to identify faults and improves the reliability of diagnosis. Attached Figure Description

[0021] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below.

[0022] Figure 1 This is a flowchart of a power distribution cable fault diagnosis method according to an embodiment of this application; Figure 2 This is a flowchart illustrating the acquisition of the target current signal according to an embodiment of this application; Figure 3 This is a flowchart illustrating the construction of the frequency scaling kernel function according to an embodiment of this application; Figure 4 This is a flowchart illustrating the construction process of the fused feature map according to an embodiment of this application; Figure 5 This is a flowchart illustrating the determination of the fault location according to an embodiment of this application; Figure 6 This is a structural block diagram of the power distribution cable fault diagnosis device according to an embodiment of this application; Figure 7 This is a schematic diagram of the structure of the electronic device provided in the embodiments of this application; Figure 8 This is a comparison diagram of the original current signal and the target current signal obtained after preprocessing in an embodiment of this application. Figure 9 This is a frequency-scaled convolutional 6-channel feature map according to an embodiment of this application; Figure 10 This is a visualization of the fusion feature t-SNE dimensionality reduction from an embodiment of this application; Figure 11 This is a comparison chart of fault classification accuracy and model parameters in an embodiment of this application. Detailed Implementation

[0023] The terms "first," "second," "third," "fourth," etc. (if present) in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" or "having" and any variations thereof are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0024] For ease of understanding, the specific process of the embodiments of the present invention will be described below. Figure 1 This is a flowchart of the distribution network cable fault diagnosis method according to an embodiment of this application. (See attached document.) Figure 1 It includes the following steps: S100: Obtain the target current signal of the target distribution cable to be diagnosed.

[0025] The target distribution network cable to be diagnosed is either a section of distribution network cable requiring fault diagnosis, or an entire distribution network cable system requiring fault diagnosis. Distribution network cables refer to cables in a power system that directly distribute and transmit electrical energy to end users such as homes, factories, and shopping malls. The target current signal is a pre-processed current signal that has had noise and other interference removed. This pre-processed current signal can be associated with the distribution network cable to be diagnosed and stored in a database. When the target current signal of the target distribution network cable needs to be obtained, its corresponding target current signal can be extracted from the database based on the identifier of the target distribution network cable.

[0026] In one example, Figure 2 For an exemplary description of obtaining the target current signal in an embodiment of this application, please refer to... Figure 2 It includes the following steps: S200: Obtain the preset sampling frequency and preset sampling duration.

[0027] S202. Based on the preset sampling frequency and preset sampling duration, determine the number of data points for the target distribution network cable to be diagnosed.

[0028] Specifically, the preset sampling frequency is as follows: It indicates that the preset sampling duration is based on This means that the specific frequency and sampling duration can be preset. Once the preset sampling frequency and duration are determined, the number of data points to be collected can be determined. ,Right now .

[0029] S204. At a selected node of the target distribution network cable to be diagnosed, the original current signal of the target distribution network cable to be diagnosed is collected by a broadband current sensor for a preset sampling time.

[0030] Specifically, this step is used to collect all current waveforms of the cable under operating conditions, including steady-state power frequency current and possible harmonics. Among them, power frequency and low-frequency harmonic current usually reflect steady-state fault characteristics, such as continuous leakage current caused by insulation aging and changes in harmonic content. The medium-frequency transient current triggered by the fault reflects the instantaneous characteristics of sudden faults, such as high-frequency current surge waves generated at the moment of short circuit or grounding.

[0031] S206. Synchronously record associated metadata during the acquisition process.

[0032] Specifically, the associated metadata includes at least the collection timestamp, sensor location identifier, and cable operating load.

[0033] S208. Logically bind the data points of the acquired raw current signal with the associated metadata in a predetermined format to generate and store the raw signal vector.

[0034] Specifically, after synchronously recording associated metadata, it can be stored as a vector according to the format of acquisition time-sensor location-operating load. .in, Represents the original current signal vector. This means that each element in the vector, i.e., the current value at each sampling point, is a real number. The dimension represents the original current signal vector, where 1 represents rows and N represents N columns. N is a positive integer indicating that the original current signal vector has a total of N elements, i.e., data from N sampling points.

[0035] S210. Preprocess the original current signal vector to obtain the target current signal.

[0036] Specifically, the purpose of preprocessing is to remove interference and purify the signal from the original acquired signal. In this example, preprocessing may involve sequentially removing slowly changing load drift through linear fitting, removing fixed power frequency interference based on adaptive notch filtering, and removing random electromagnetic noise through wavelet threshold filtering.

[0037] It is understandable that preprocessing, in a physical sense, is performed on the current signal, while in the specific algorithm and code implementation, it is performed on the data vector Xraw representing that signal. They are different levels of representation of the same thing: the former is the physical essence of the latter, and the latter is the digital carrier of the former. Therefore, after preprocessing the original current signal vector, the target current signal can be obtained.

[0038] S102. Based on the target sampling frequency of the target current signal, initialize the frequency scaling factor and center frequency of the Morlet wavelet basis function, and construct the frequency scaling kernel function.

[0039] Specifically, Figure 3 This is an exemplary description of constructing a frequency scaling kernel function according to the application embodiments, see below. Figure 3 It includes the following steps: S300, the target sampling frequency for acquiring the target current signal.

[0040] Specifically, the target sampling frequency of the target current signal can be the preset sampling frequency set when acquiring the original current signal. .

[0041] S302. Based on the target sampling frequency, determine the kernel center frequency and the range of time sample points for the Morlet wavelet basis function.

[0042] Specifically, the core center frequency The range of values ​​for the time sample points is: ; In the formula, For the core center frequency, The core center frequency is in frequency form. The target sampling frequency is the target current signal.

[0043] S304. Based on the core harmonic frequency band of the target type of cable fault, set a set of initial trainable scaling factors.

[0044] In this example, the core harmonic frequency band of the target type can be one of six core harmonic frequency bands.

[0045] S306. Construct a frequency scaling kernel function using the kernel center frequency, time sample points, and an initialized trainable scaling factor.

[0046] Specifically, the expression for the frequency scaling kernel function is: ; In the formula, : Frequency scaling kernel output value; Time sample points, from the first step Confirmed, covering one complete fault transient cycle; : Trainable scaling factor, the initial trainable scaling factor is , The core channel number is used to match the target type of cable fault in the core harmonic frequency band; Bandwidth constant.

[0047] S104. Perform one-dimensional convolution operation on the target current signal based on the frequency scaling kernel function to generate a multi-channel fault feature map.

[0048] Specifically, convolution stride , The expression for one-dimensional convolution is: ; In the formula: : Convolution stride; Fault feature map output by convolution, dimension , For the number of channels, The length of the feature map is given, and each channel corresponds to a type of current fault harmonic feature. : Feature map time axis index, value This corresponds to the time window after downsampling; The kernel function length is matched with the kernel parameters in the second step to ensure the capture of transient local features of the current. : No. Each time sample point, scaling factor The kernel value comes from the initialization result in the second step and is optimized with fault characteristics in subsequent training; The amplitude of the preprocessed current signal at the corresponding index comes from step two. ; Batch normalization avoids imbalance in weight updates caused by differences in current amplitude; Preprocessed current signal The total number of sampling points; Feature map length The calculation formula is: .

[0049] S106. Process the fault feature map to obtain a fused feature map.

[0050] In one example, Figure 4 This refers to the process of constructing the fused feature map; see [link / reference]. Figure 4 It includes the following steps: S400. Perform one-dimensional convolution and pooling on the fault feature map to obtain a time-domain refined feature map.

[0051] S402. Reshape the time-domain refined feature map into a two-dimensional tensor, and perform a two-dimensional convolution operation on the two-dimensional tensor to obtain a fused feature map.

[0052] S108. Input the fused feature map into the target connected linear layer for processing to determine the fault type corresponding to the fused feature map.

[0053] The dimension of the weight matrix of the target connected linear layer is configured to be associated with the frequency channel dimension of the fused feature map, without changing with the time point of the fused feature map.

[0054] Specifically, after the fused feature map is input into the target and connected to the linear layer, classification scoring is performed based on the scoring formula to obtain the fault classification scores corresponding to each of the multiple fault types in the fused feature map. The scores of multiple fault classifications are normalized to obtain the fault type corresponding to the fused feature map. The scoring formula is as follows: ; In the formula, : No. Classification scores for different types of faults ; : Feature map channel index, fixed at 1, from the single-channel output of the fused feature map; Time axis index, value ; : Length of the time axis after downsampling; Frequency channel index, value ; : The transpose of the matrix in The transpose of the fused feature map whose position is determined by the elements; Target connection weights of the linear layer, dimension ; : No. Fault bias term.

[0055] S110. Determine the location of the fault point based on the time difference of the peak values ​​of fault harmonics and the propagation speed of current waves at adjacent acquisition points in the fault feature map.

[0056] Specifically, Figure 5 This is a flowchart illustrating the determination of the fault location according to an embodiment of this application. (See attached document.) Figure 5 It includes the following steps: S500: Extract the peak time of the first harmonic current fault and the peak time of the second harmonic current fault from the fault feature map of adjacent acquisition points.

[0057] S502. Determine the current wave propagation speed based on the type of the target distribution network cable to be diagnosed.

[0058] S504. Based on the peak time of the first harmonic of the current fault, the peak time of the second harmonic of the current fault, and the propagation speed of the current wave, the location of the fault point is determined by combining the distance formula.

[0059] The distance formula is as follows: ; In the formula: : The distance from the fault point to the starting data collection point; Spacing between data collection points; : Speed ​​of propagation of electric current waves; : The peak time of the first harmonic; : The peak time of the second harmonic; The absolute value of the time difference between the peak time of the first harmonic and the peak time of the second harmonic.

[0060] S112. Based on the fault type, fault location, and distribution network topology, determine the fault target investigation path, and investigate the target distribution network cable to be diagnosed based on the fault target investigation path, so as to diagnose the target fault of the target distribution network cable to be diagnosed.

[0061] Specifically, based on the distribution network structure, including cable laying paths and maintenance well locations, risk levels are classified according to Type*, such as Level 1: phase-to-phase short circuit / open circuit; Level 2: single-phase grounding / lightning strike; Level 3: inter-turn fault / poor contact. Level 1 faults prioritize direct access via maintenance wells, while Level 2 / 3 faults prioritize easily accessible paths. Joints within 50m before and after the fault point at key nodes along the path are marked, and corresponding troubleshooting operations are pushed, such as checking the status of joints near short-circuit fault points. A visual path map is output to the field terminal to achieve rapid fault diagnosis, thereby ultimately determining the target fault of the distribution network cable to be diagnosed.

[0062] For example, to facilitate understanding of this solution, please refer to the appendix. Figures 8 to 11 For example: First, the raw current signal is acquired. Specifically, the acquisition of the raw current signal includes the acquisition of transient and steady-state components. Because faults in distribution network cables, such as short circuits, grounding, and inter-turn faults, directly cause current waveform distortion and generate characteristic harmonics, key characteristic signals are captured at high-fault nodes such as cable terminals, intermediate joints, and branch boxes: power frequency and low-frequency harmonic currents, as well as fault-triggered medium-frequency transient currents, by setting the sampling rate. Collection time Number of data points per sample Simultaneously record the cable operating load at the time of data collection. The collected current signal Stored as a vector in the format of acquisition time-sensor location-operating load. .

[0063] Further preprocessing of the current signal and initialization of the frequency scaling kernel are carried out. Specifically, the signal-to-noise ratio is improved through multi-stage preprocessing to address power frequency interference, electromagnetic noise and baseline drift in the acquired current signal. At the same time, the frequency scaling kernel is initialized based on the harmonic characteristics of the current fault, laying the foundation for extracting fault features.

[0064] First of all Linear fitting is used to eliminate the influence of load drift, then adaptive notch filtering is used to remove power frequency interference, and finally wavelet threshold filtering is used to remove electromagnetic noise, resulting in the preprocessed current signal. .

[0065] right Normalization is performed to prevent amplitude differences from affecting subsequent convolution weight updates. Following the previous step... The frequency scaling kernel is designed using Morlet basis functions, and the kernel function initialization expression is: ; In the formula, : Frequency scaling kernel output value, the core operator for subsequent convolution operations; Time sample points (unit: seconds), values ,Depend on Confirmed, covering one complete fault transient cycle; : Trainable scaling factor, the initial trainable scaling factor is , The number of core channels is used to match the six core harmonic frequency bands of cable faults. Nuclear center frequency, ; In the formula, For the core center frequency, The core center frequency is in frequency form. The target sampling frequency for the target current signal; : Bandwidth constant, ensuring the core's frequency resolution remains stable under different scaling factors.

[0066] Specifically, in this scheme, the operation of constructing the frequency scaling kernel function corresponds to a one-dimensional convolutional layer, whose input dimension can be the target current signal, and the dimension is... Where U is the signal length, and its output is , dimension Where L is the kernel function length, C is the number of kernel channels, and the range of values ​​for the time sample point z is... After sampling, multiple discrete time points are obtained. Therefore, the kernel function length L is determined by the transient period duration (ensuring coverage of one period).

[0067] The output of the frequency scaling kernel function is concatenated with the preprocessed current signal through a one-dimensional convolution operation to generate a fault feature map. The fault feature map is a multi-channel, single-time-dimensional feature representation, which can be further used to generate the final fused feature map through "time-frequency fusion" (such as splicing or attention mechanisms).

[0068] like Figure 8 The figure shows a comparison between the original current signal and the target current signal obtained after preprocessing in an embodiment of this application, used to verify the effectiveness of the current signal preprocessing. Figure 8The upper curve represents the original current signal, which includes a 50Hz power frequency component, random electromagnetic noise, and the third harmonic caused by the fault. The signal amplitude fluctuates chaotically, and the fault characteristics are masked by noise. The lower curve represents the preprocessed signal after detrending, adaptive notch filtering, and wavelet threshold filtering. Power frequency interference and electromagnetic noise are effectively suppressed, and the transient impact of the fault and the third harmonic characteristics are clearly highlighted. This proves that the preprocessing process can accurately retain the core fault information and lay the foundation for subsequent feature extraction.

[0069] Furthermore, frequency scaling convolution is performed to extract current fault features: Specifically, by using trainable frequency-scalable convolutional layers, time-frequency domain features are extracted from the preprocessed current signal. For the differences in current harmonics of different faults, the convolutional kernels adaptively adjust the scaling factor to separate fault features from normal current components. At the same time, downsampling reduces the computational load.

[0070] In practice, the data obtained in the previous step will be used as a basis for implementation. and Perform 1D convolution operation, convolution stride , The calculation formula is: ; In the formula: : Convolution stride; Fault feature map output by convolution, dimension ( For the number of channels, (where the length is the feature map length), each channel corresponds to a type of current fault harmonic feature; : Feature map time axis index, value This corresponds to the time window after downsampling; The kernel function length is matched with the kernel parameters in the second step to ensure the capture of transient local features of the current. : No. Each time sample point, scaling factor The kernel value comes from the initialization result in the second step and is optimized with fault characteristics in subsequent training; The amplitude of the preprocessed current signal at the corresponding index comes from step two. ; Batch normalization avoids imbalance in weight updates caused by differences in current amplitude; Preprocessed current signal The total number of sampling points; Feature map length The calculation formula is: .

[0071] Specifically, in this scheme, the operation of constructing the frequency scaling kernel function corresponds to a one-dimensional convolutional layer, whose input and output dimensions and functions are as follows: The input is the target current signal, with dimensions [U, 1], meaning the input is a single-channel current signal with a signal length of U; the output is a multi-channel frequency scaling kernel function, with dimensions [L, C], where L is the kernel function length, which is related to the range of time sample points. The number of sampling points is consistent, by It is determined that L×sampling interval = 4π / This means covering a complete fault transient cycle. C is the number of core channels, which is consistent with the number of core harmonic frequency bands of the cable fault. For example, when there are 6 core harmonic frequency bands, C=6.

[0072] Furthermore, the output of the frequency scaling kernel function is subjected to a series of convolution operations with the preprocessed current signal. Connect and generate fault feature maps Its convolutional budget logic is as follows: Input 1: The dimension is [L,C], and the input is the preprocessed current signal. Dimensions [N, 1], Output: Fault Feature Map The dimension is [P, C], where P is the length of the feature map, derived from the formula... calculate.

[0073] Furthermore, Each channel, for current signal Perform locally weighted convolution, i.e., the kernel at the j-th time sample point. and The Multiplying the sample points together and summing them up, we get the nth sample point. The first time window, the... The characteristic values ​​of the channel.

[0074] Furthermore, batch normalization is performed. The feature values ​​of each time window are normalized to ensure that the feature amplitudes of different channels are consistent, thus avoiding imbalance in weight updates during training.

[0075] Furthermore, fault characteristic diagrams It is a multi-channel, single-time-dimensional feature representation. The final fused feature map can be generated through time-frequency fusion, and its dimension is typically [missing information]. ,in, This represents the length of the time axis after downsampling. This represents the number of channels after merging.

[0076] like Figure 9 The image shown is a frequency-scaled convolutional 6-channel feature map according to an embodiment of this application. Figure 9 The output results of frequency scaling convolution for extracting current fault features are shown. Figure 9 The six sub-plots of channels 1 to 6 correspond to the six channels of the frequency scaling kernel, adapting to the core harmonic frequency bands of six types of faults. Each channel focuses on fault characteristics in a different frequency range: channels 1-2 are in the low-frequency band, highlighting the steady-state characteristics of the power frequency and the third harmonic, with the curves exhibiting regular periodic fluctuations, corresponding to the characteristic differences between healthy states and minor faults; channels 3-4 are in the mid-frequency band, capturing the amplitude changes of transient impacts during faults, with a significant peak appearing during the 0.3-0.35s fault period, reflecting the core information of transient currents; channels 5-6 are in the mid-to-high frequency band, enhancing the distortion characteristics of fault harmonics, and the curves after filtering out residual interference more clearly reflect the specificity of the fault type. The features of the six channels are complementary, proving that frequency scaling convolution can adaptively separate fault information in different frequency dimensions, providing high-quality feature input for subsequent spatiotemporal fusion.

[0077] Further, spatiotemporal current feature fusion (1D+2D convolution) is carried out. Specifically, the time-domain dynamics and frequency-domain specificity of current fault signals are refined by 1D convolution and spatiotemporal correlation is fused by 2D convolution to uncover the fault modes hidden in the current signals and avoid misjudgment caused by single-dimensional features.

[0078] The first phase will Each frequency channel is independently input into a 1D convolutional layer, which uses a 3×1 kernel to capture local temporal features of the current, and reduces redundancy through 2×1 max pooling. The computation formula is as follows: ; This is a max pooling operation; : Input feature map of the (t-1)th layer of the network; : No. 1D convolution output; 1D convolution can train weights, dimensions Adaptive updates based on current fault characteristics; Bias term, used to compensate for the shift in the convolution mean; Time feature map length .

[0079] The second phase will Remodeling into 2D tensors Where rows = frequency channels, columns = time axes, input is a 3×3 single-channel 2D convolutional layer, and ReLU activation is used to enhance the nonlinear expression, as shown in the formula: ; in, This is a time-domain refined feature map after one-dimensional convolution; : Fusing feature maps, dimensions The complete spatiotemporal characteristics of current faults; 2D convolution weights, dimensions The remaining parameters are the same as those for 1D convolution.

[0080] For example, Figure 10 This is a t-SNE dimensionality reduction visualization of the fused features according to an embodiment of this application. The figure shows the t-SNE dimensionality reduction visualization result of the fused spatiotemporal current features, verifying the category discrimination capability of the fused features. The figure includes seven types of faults: healthy, single-phase grounding, phase-to-phase short circuit, inter-turn fault, open circuit, lightning strike, and poor contact. Each type of fault is represented by a different color and marker. Figure 10 As can be seen, faults of the same category form compact clusters, while clusters of different categories have clear boundaries and no significant overlap: healthy sample clusters are concentrated in the lower left corner, short-circuit faults such as phase-to-phase short circuits and open circuits are clustered in the upper right corner, and minor faults such as inter-turn faults and poor contact clusters, although relatively close, can still be clearly distinguished. This result demonstrates that the process of refining temporal features through 1D convolution and fusing spatiotemporal correlations through 2D convolution can effectively uncover hidden fault modes in current signals. The generated fused features have strong discriminative power, providing reliable support for subsequent fault classification.

[0081] Furthermore, current fault classification is achieved based on a partially connected linear layer. Specifically, given the limited computing power of distribution network field equipment, a partially connected linear layer replaces the fully connected layer, significantly reducing the number of parameters. Simultaneously, fault types are accurately determined based on fused features. Specifically, this involves... The direct input part connects to a 2D linear layer, and the calculation formula is: ; : No. Classification scores for different types of faults The seven types of faults include health, single-phase grounding, phase-to-phase short circuit, turn-to-turn fault, open circuit, lightning strike, and poor contact. Feature map channel index, fixed at 1 (from Single-channel output); Time axis index, value (From step four) ); Length of the time axis after downsampling; Frequency channel index, value (From step 3) ) : The transpose of the matrix in The transpose of the fused feature map whose position is determined by the elements; Partial connection weights, dimensions It is only associated with frequency channels, reducing computing power requirements; : No. The fault bias term improves the accuracy of fault classification in small samples.

[0082] Subsequently Perform Softmax normalization and select the highest-scoring category with a confidence level ≥ 90% as the fault type. .

[0083] like Figure 11 The figure shows a comparison of fault classification accuracy and model parameters in an embodiment of this application. This figure compares the number of trainable parameters and fault classification accuracy of the present invention with existing deep learning models, highlighting the lightweight advantage brought by the partially connected linear layers (target connection layers). The bar chart represents the number of trainable parameters, and the line chart represents the classification accuracy (right vertical axis). In existing technologies, ResNet and LSTM-RegNet models have millions of parameters, achieving accuracy close to 99% but requiring extremely high computational power; fully connected CNNs have approximately 46,000 parameters with an accuracy of only 98.5%; while the model of the present invention, through frequency-scaling convolution combined with partially connected linear layers, has only 644 parameters and an accuracy exceeding 98.5%, significantly outperforming existing technologies. This figure intuitively verifies the dual advantages of the present invention in terms of lightweight design and high accuracy, solving the core problems of bulky and computationally demanding existing models, and meeting the actual needs of portable diagnostic equipment in power distribution networks.

[0084] Furthermore, fault location and troubleshooting path generation are performed based on current characteristics. Specifically, by combining the time difference and wave propagation characteristics of current faults, the fault location is accurately calculated, and the optimal troubleshooting path is generated based on the fault type. Since the current wave propagates at a stable speed in the cable, the time difference between the harmonic peak values ​​of current faults at adjacent sampling points can be directly used for location.

[0085] It consists of two steps: 1. Fault Location Calculation: Extract the peak times t1 and t2 of the current fault harmonics from adjacent sampling points (distance d) in the fault feature map. Determine the current wave propagation velocity v based on the cable type. The calculation formula is as follows: ; : Distance from the fault point to the starting data collection point (unit: m); Spacing between data collection points (unit: m); Current wave propagation speed (unit: m / s); : The peak time of the first harmonic; : The peak time of the second harmonic; The absolute value of the time difference between the peak time of the first harmonic and the peak time of the second harmonic.

[0086] Based on the power distribution network structure, including cable laying paths and maintenance well locations, risk levels are categorized by Type*, such as Level 1: phase-to-phase short circuit / open circuit; Level 2: single-phase grounding / lightning strike; Level 3: inter-turn fault / poor contact. Level 1 faults prioritize direct access via maintenance wells, while Level 2 / 3 faults prioritize easily accessible routes. Key nodes in the path are marked, such as joints within 50m before and after the fault point, and corresponding troubleshooting operations are pushed, such as checking the status of joints near the short-circuit fault point. A visual path map is output to the field terminal to achieve rapid fault troubleshooting.

[0087] Understandably, this technical solution is adaptable to different laying scenarios such as underground and overhead, as well as various fault types such as grounding, short circuit, and inter-turn faults. It is highly practical for engineering and easy to promote and apply in distribution network operation and maintenance, effectively reducing power outage losses caused by distribution network faults.

[0088] Furthermore, Figure 6 The above is a structural block diagram of the distribution network cable fault diagnosis device according to the application embodiment, such as... Figure 6 As shown, the device includes: The target current signal acquisition module is used to acquire the target current signal of the target distribution network cable to be diagnosed. The frequency scaling kernel function construction module is used to initialize the frequency scaling factor and center frequency of the Morlet wavelet basis function based on the target sampling frequency of the target current signal, and construct the frequency scaling kernel function. The fault feature map generation module is used to perform one-dimensional convolution operation on the target current signal based on the frequency scaling kernel function to generate a multi-channel fault feature map. The fusion feature map determination module is used to process the fault feature map to obtain the fusion feature map; The fault type determination module is used to input the fused feature map into the target connected linear layer for processing in order to determine the fault type corresponding to the fused feature map. The dimension of the weight matrix of the target connected linear layer is configured to be related to the frequency channel dimension of the fused feature map, without changing with the time point of the fused feature map. The fault location determination module determines the fault location based on the time difference of the fault harmonic peak values ​​of adjacent acquisition points in the fault feature map and the current wave propagation speed. The target fault diagnosis module is used to determine the fault target investigation path based on the fault type, fault location and distribution network topology, and to investigate the target distribution network cable to be diagnosed based on the fault target investigation path, so as to diagnose the target fault of the target distribution network cable.

[0089] The application of the relevant modules of the device in this example can be referred to the relevant introduction of the method principle above, and will not be repeated here.

[0090] above Figure 6 The distribution network cable fault diagnosis device in this embodiment of the invention will be described in detail from the perspective of modular functional entities. The electronic equipment in this embodiment of the invention will be described in detail from the perspective of hardware processing.

[0091] Figure 7 This is a schematic diagram of the structure of an electronic device 700 provided in an embodiment of the present invention. The electronic device 700 can vary significantly due to differences in configuration or performance, and may include one or more central processing units (CPUs) 710 (e.g., one or more processors) and a memory 720, and one or more storage media 730 (e.g., one or more mass storage devices) for storing application programs 733 or data 732. The memory 720 and storage media 730 can be temporary or persistent storage. The program stored in the storage media 730 may include one or more modules (not shown in the diagram), each module including a series of instruction operations on the electronic device 700. Furthermore, the processor 710 may be configured to communicate with the storage media 730 and execute the series of instruction operations in the storage media 730 on the electronic device 700.

[0092] Electronic device 700 may also include one or more power supplies 740, one or more wired or wireless network interfaces 750, one or more input / output interfaces 760, and / or one or more operating systems 731, such as Windows Server, MacOSX, Unix, Linux, FreeBSD, etc. Those skilled in the art will understand that... Figure 7The illustrated electronic device structure does not constitute a limitation on electronic devices and may include more or fewer components than illustrated, or combine certain components, or have different component arrangements.

[0093] The present invention also provides a computer-readable storage medium, which can be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium, wherein the computer-readable storage medium stores instructions that, when executed on a computer, cause the computer to perform the steps of any of the above-described distribution network cable fault diagnosis methods.

[0094] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the system, device, or unit described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0095] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0096] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for diagnosing faults in distribution network cables, characterized in that, The method includes: Acquire the target current signal of the distribution network cable to be diagnosed; Based on the target sampling frequency of the target current signal, initialize the frequency scaling factor and center frequency of the Morlet wavelet basis function, and construct the frequency scaling kernel function; The target current signal is subjected to a one-dimensional convolution operation based on the frequency scaling kernel function to generate a multi-channel fault feature map. The fault feature map is processed to obtain a fused feature map; The fused feature map is input into a target connected linear layer for processing to determine the fault type corresponding to the fused feature map. The dimension of the weight matrix of the target connected linear layer is configured to be associated with the frequency channel dimension of the fused feature map, without changing with the time point of the fused feature map. The location of the fault point is determined based on the time difference of the peak fault harmonics and the propagation speed of the current wave at adjacent acquisition points in the fault feature map. Based on the fault type, the location of the fault point, and the distribution network topology, a fault target investigation path is determined, and the target distribution network cable to be diagnosed is investigated based on the fault target investigation path to diagnose the target fault of the target distribution network cable.

2. The method for diagnosing distribution network cable faults according to claim 1, characterized in that, The acquisition of the target current signal of the target distribution network cable to be diagnosed includes: Obtain the preset sampling frequency and preset sampling duration; Based on the preset sampling frequency and the preset sampling duration, the number of data points of the target distribution network cable to be diagnosed is determined; At a selected node of the target distribution network cable to be diagnosed, the original current signal of the target distribution network cable to be diagnosed is collected by a broadband current sensor for the preset sampling duration; Synchronously record associated metadata during the acquisition process, including at least the acquisition timestamp, sensor location identifier, and cable operating load; The data points of the acquired raw current signal are logically bound to the associated metadata in a predetermined format to generate and store the raw current signal vector. The original current signal vector is preprocessed to obtain the target current signal.

3. The method for diagnosing distribution network cable faults according to claim 1, characterized in that, The process of initializing the frequency scaling factor and center frequency of the Morlet wavelet basis function based on the target sampling frequency of the target current signal, and constructing the frequency scaling kernel function includes: Obtain the target sampling frequency of the target current signal; Based on the target sampling frequency, determine the kernel center frequency and the range of time sample points for the Morlet wavelet basis function; Based on the core harmonic frequency band of the target type of cable fault, set a set of initial trainable scaling factors. The frequency scaling kernel function is constructed using the kernel center frequency, the time sample points, and the initialized trainable scaling factor; Among them, the core center frequency ; In the formula, For the core center frequency, The core center frequency is in frequency form. The target sampling frequency for the target current signal; The range of values ​​for time sample points is ; The expression for the frequency scaling kernel function is: ； In the formula, : Frequency scaling kernel output value; Time sample points, from the first step Confirmed, covering one complete fault transient cycle; : Trainable scaling factor, the initial trainable scaling factor is , The core channel number is used to match the target type of cable fault in the core harmonic frequency band; Bandwidth constant.

4. The method for diagnosing distribution network cable faults according to claim 1, characterized in that, The target current signal is convolved using the frequency scaling kernel function to generate a multi-channel fault feature map, with the convolution stride being... , The expression for one-dimensional convolution is: ； In the formula: : Convolution stride; Fault feature map output by convolution, dimension , For the number of channels, The length of the feature map is given, and each channel corresponds to a type of current fault harmonic feature. : Feature map time axis index, value This corresponds to the time window after downsampling; The kernel function length is matched with the kernel parameters in the second step to ensure the capture of transient local features of the current. : No. Each time sample point, scaling factor The kernel value comes from the initialization result in the second step and is optimized with fault characteristics in subsequent training; The amplitude of the preprocessed current signal at the corresponding index comes from step two. ; Batch normalization avoids imbalance in weight updates caused by differences in current amplitude; Preprocessed current signal The total number of sampling points; Feature map length The calculation formula is: 。 5. The method for diagnosing faults in distribution network cables according to claim 1, characterized in that, The process of processing the fault feature map to obtain the fused feature map includes: The fault feature map is subjected to one-dimensional convolution and pooling to obtain a time-domain refined feature map; The temporal refined feature map is reshaped into a two-dimensional tensor, and a two-dimensional convolution operation is performed on the two-dimensional tensor to obtain the fused feature map.

6. The method for diagnosing distribution network cable faults according to claim 1, characterized in that, The step of inputting the fused feature map into the target connected linear layer for processing to determine the fault type corresponding to the fused feature map includes: After the fused feature map is input into the target connected linear layer, classification scoring is performed based on the scoring formula to obtain the fault classification scores corresponding to each of the multiple fault types in the fused feature map. The scores of multiple fault classifications are normalized to obtain the fault type corresponding to the fused feature map. The scoring formula is as follows: ； In the formula, : No. Classification scores for different types of faults ; : Feature map channel index, fixed at 1, from the single-channel output of the fused feature map; Time axis index, value ; : Length of the time axis after downsampling; Frequency channel index, value ; : The transpose of the matrix in The transpose of the fused feature map whose position is determined by the elements; Target connection weights of the linear layer, dimension ; : No. Fault bias term.

7. The method for diagnosing distribution network cable faults according to claim 1, characterized in that, The method of determining the location of the fault point based on the time difference of the peak values ​​of fault harmonics and the propagation velocity of current waves at adjacent acquisition points in the fault feature map includes: Extract the peak times of the first harmonic current fault and the peak times of the second harmonic current fault from adjacent acquisition points in the fault feature map. Determine the current wave propagation speed based on the type of the target distribution network cable to be diagnosed; Based on the peak times of the first harmonic of the current fault, the peak times of the second harmonic of the current fault, and the propagation speed of the current wave, the location of the fault point is determined using a distance formula, wherein the distance formula is: ； In the formula: : The distance from the fault point to the starting data collection point; Spacing between data collection points; : Speed ​​of propagation of electric current waves; : The peak time of the first harmonic; : The peak time of the second harmonic; The absolute value of the time difference between the peak time of the first harmonic and the peak time of the second harmonic.

8. A fault diagnosis device for distribution network cables, characterized in that, The device includes: The target current signal acquisition module is used to acquire the target current signal of the target distribution network cable to be diagnosed. The frequency scaling kernel function construction module is used to initialize the frequency scaling factor and center frequency of the Morlet wavelet basis function based on the target sampling frequency of the target current signal, and construct the frequency scaling kernel function. The fault feature map generation module is used to perform a one-dimensional convolution operation on the target current signal based on the frequency scaling kernel function to generate a multi-channel fault feature map. The fusion feature map determination module is used to process the fault feature map to obtain the fusion feature map; The fault type determination module is used to input the fused feature map into the target connected linear layer for processing in order to determine the fault type corresponding to the fused feature map. The dimension of the weight matrix of the target connected linear layer is configured to be associated with the frequency channel dimension of the fused feature map, and does not change with the time point of the fused feature map. The fault location determination module determines the fault location based on the time difference of the fault harmonic peak values ​​of adjacent acquisition points and the current wave propagation speed in the fault feature map. The target fault diagnosis module is used to determine the fault target investigation path based on the fault type, the fault point location and the distribution network topology, and to investigate the target distribution network cable to be diagnosed based on the fault target investigation path, so as to diagnose the target fault of the target distribution network cable to be diagnosed.

9. An electronic device, characterized in that, The electronic device includes a memory and at least one processor, the memory storing instructions; the at least one processor invokes the instructions in the memory to cause the electronic device to perform the steps of the power distribution cable fault diagnosis method as described in any one of claims 1-7.

10. A computer-readable storage medium storing instructions thereon, characterized in that, When the instructions are executed by the processor, they implement the various steps of the power distribution cable fault diagnosis method as described in any one of claims 1-7.

Images